Abstracting spreadsheet data flow through hypergraph redrawing
We believe the error prone nature of traditional spreadsheets is due to their low level of abstraction. End user programmers are forced to construct their data models from low level cells which we define as "a data container or manipulator linked by user-intent to model their world and positioned to reflect its structure". Spreadsheet cells are limited in what they may contain (scalar values) and the links between them are inherently hidden. This paper proposes a method of raising the level of abstraction of spreadsheets by "redrawing the boundary" of the cell. To expose the hidden linkage structure we transform spreadsheets into fine-grained graphs with operators and values as nodes. "cells" are then represented as hypergraph edges by drawing a boundary "wall" around a set of operator/data nodes. To extend what cells may contain and to create a higher level model of the spreadsheet we propose that researchers should seek techniques to redraw these boundaries to create higher level "cells" which will more faithfully represent the end-user's real world/mental model. We illustrate this approach via common sub-expression identification and the application of sub-tree isomorphisms for the detection of vector (array) operations.